Generalizing trajectories¶

Binder IPYNB HTML

To reduce the size (number of points) of trajectory objects, we can generalize them, for example, using:

  • Spatial generalization, such as Douglas-Peucker algorithm
  • Temporal generalization by down-sampling, i.e. increasing the time interval between records
  • Spatiotemporal generalization, e.g. using Top-Down Time Ratio algorithm

Documentation

A closely related type of operation is trajectory smoothing which is coverd in a separate notebook.

In [1]:
import pandas as pd
import geopandas as gpd
import movingpandas as mpd
import shapely as shp
import hvplot.pandas 
import matplotlib.pyplot as plt

from geopandas import GeoDataFrame, read_file
from shapely.geometry import Point, LineString, Polygon
from datetime import datetime, timedelta
from holoviews import opts

import warnings
warnings.filterwarnings('ignore')

plot_defaults = {'linewidth':5, 'capstyle':'round', 'figsize':(9,3), 'legend':True}
opts.defaults(opts.Overlay(active_tools=['wheel_zoom'], frame_width=500, frame_height=400))

mpd.show_versions()

MovingPandas 0.16.0

SYSTEM INFO
-----------
python     : 3.10.8 | packaged by conda-forge | (main, Nov 22 2022, 08:16:53) [MSC v.1929 64 bit (AMD64)]
executable : H:\miniconda3\envs\mpd-ex\python.exe
machine    : Windows-10-10.0.19045-SP0

GEOS, GDAL, PROJ INFO
---------------------
GEOS       : None
GEOS lib   : None
GDAL       : 3.5.0
GDAL data dir: None
PROJ       : 9.0.0
PROJ data dir: H:\miniconda3\pkgs\proj-9.0.0-h1cfcee9_1\Library\share\proj

PYTHON DEPENDENCIES
-------------------
geopandas  : 0.13.0
pandas     : 2.0.1
fiona      : 1.8.21
numpy      : 1.23.5
shapely    : 1.8.2
rtree      : 1.0.1
pyproj     : 3.3.1
matplotlib : 3.7.1
mapclassify: 2.4.3
geopy      : 2.3.0
holoviews  : 1.14.9
hvplot     : 0.8.3
geoviews   : 1.9.6
stonesoup  : 0.1b12
In [2]:
gdf = read_file('../data/geolife_small.gpkg')
traj_collection = mpd.TrajectoryCollection(gdf, 'trajectory_id', t='t')
In [3]:
original_traj = traj_collection.trajectories[1]
print(original_traj)

Trajectory 2 (2009-06-29 07:02:25 to 2009-06-29 11:13:12) | Size: 897 | Length: 38764.6m
Bounds: (116.319212, 39.971703, 116.592616, 40.082514)
LINESTRING (116.590957 40.071961, 116.590905 40.072007, 116.590879 40.072027, 116.590915 40.072004,
In [4]:
original_traj.plot(column='speed', vmax=20, **plot_defaults)
Out[4]:
<Axes: >

Spatial generalization (DouglasPeuckerGeneralizer)¶

Try different tolerance settings and observe the results in line geometry and therefore also length:

In [5]:
dp_generalized  = mpd.DouglasPeuckerGeneralizer(original_traj).generalize(tolerance=0.001)
dp_generalized.plot(column='speed', vmax=20, **plot_defaults)
Out[5]:
<Axes: >
In [6]:
dp_generalized
Out[6]:
Trajectory 2 (2009-06-29 07:02:25 to 2009-06-29 11:13:12) | Size: 31 | Length: 36921.9m
Bounds: (116.319709, 39.971775, 116.592616, 40.082369)
LINESTRING (116.590957 40.071961, 116.590367 40.073957, 116.590367 40.073957, 116.590367 40.073957,
In [7]:
print('Original length: %s'%(original_traj.get_length()))
print('Generalized length: %s'%(dp_generalized.get_length()))

Original length: 38764.575482545886
Generalized length: 36921.91845209718

Temporal generalization (MinTimeDeltaGeneralizer)¶

An alternative generalization method is to down-sample the trajectory to ensure a certain time delta between records:

In [8]:
time_generalized = mpd.MinTimeDeltaGeneralizer(original_traj).generalize(tolerance=timedelta(minutes=1))
time_generalized.plot(column='speed', vmax=20, **plot_defaults)
Out[8]:
<Axes: >
In [9]:
time_generalized.to_point_gdf().head(10)
Out[9]:
id sequence trajectory_id tracker geometry
t
2009-06-29 07:02:25 1556 1090 2 0 POINT (116.59096 40.07196)
2009-06-29 07:03:25 1569 1103 2 0 POINT (116.59069 40.07225)
2009-06-29 07:04:25 1582 1116 2 0 POINT (116.59037 40.07396)
2009-06-29 07:05:25 1595 1129 2 0 POINT (116.59260 40.07411)
2009-06-29 07:06:25 1610 1144 2 0 POINT (116.59258 40.07420)
2009-06-29 07:07:25 1623 1157 2 0 POINT (116.59235 40.07602)
2009-06-29 07:08:25 1635 1169 2 0 POINT (116.58939 40.07794)
2009-06-29 07:09:25 1647 1181 2 0 POINT (116.58911 40.08171)
2009-06-29 07:10:25 1659 1193 2 0 POINT (116.58829 40.08232)
2009-06-29 07:11:25 1672 1206 2 0 POINT (116.58689 40.08230)
In [10]:
original_traj.to_point_gdf().head(10)
Out[10]:
id sequence trajectory_id tracker geometry
t
2009-06-29 07:02:25 1556 1090 2 0 POINT (116.59096 40.07196)
2009-06-29 07:02:30 1557 1091 2 0 POINT (116.59091 40.07201)
2009-06-29 07:02:35 1558 1092 2 0 POINT (116.59088 40.07203)
2009-06-29 07:02:40 1559 1093 2 0 POINT (116.59091 40.07200)
2009-06-29 07:02:45 1560 1094 2 0 POINT (116.59096 40.07198)
2009-06-29 07:02:50 1561 1095 2 0 POINT (116.59101 40.07196)
2009-06-29 07:02:55 1562 1096 2 0 POINT (116.59099 40.07198)
2009-06-29 07:03:00 1563 1097 2 0 POINT (116.59098 40.07199)
2009-06-29 07:03:05 1564 1098 2 0 POINT (116.59097 40.07200)
2009-06-29 07:03:10 1565 1099 2 0 POINT (116.59097 40.07200)

Spatiotemporal generalization (TopDownTimeRatioGeneralizer)¶

In [11]:
tdtr_generalized = mpd.TopDownTimeRatioGeneralizer(original_traj).generalize(tolerance=0.001)

Let's compare this to the basic Douglas-Peucker result:

In [12]:
fig, axes = plt.subplots(nrows=1, ncols=2, figsize=(19,4))
tdtr_generalized.plot(ax=axes[0], column='speed', vmax=20, **plot_defaults)
dp_generalized.plot(ax=axes[1], column='speed', vmax=20, **plot_defaults)
Out[12]:
<Axes: >
In [ ]: